ISSUE-104: Several issues with the current treatment of disallowed vocabulary

Hi!

At the last telco, Boris (cc'ed) told me that ISSUE-104 (disallowed 
vocabulary) is probably moot, since there is already a notion of a "reserved 
vocabulary" in sec. 2.2 of the DL-Syntax document [1].

I have read this section in the meanwhile, and have found three issues, which 
still remain. So I disagree that this issue is already resolved. The three 
issues are pretty independent of each other, and can therefore be discussed 
separately.

I claim:

  (A) Backwards Compatibility:
      The set of disallowed vocabulary covers URIs,
      which were allowed to be used in OWL 1 DL.
      Hence, we have a backwards-compatibility issue.

  (B) Responsibility:
      The disallowed vocabulary should be in the
      RDF mapping, not in the DL-syntax document.

  (C) Relaxation:
      It seems to be possible that a fraction of the
      previously disallowed vocabulary can be allowed
      to be used in OWL 2 DL.

I will provide a more detailed argumentation for all these claims below.

(A): Backwards compatibility
----------------------------

The current disallowed (or "reserved") vocabulary is defined to be the set of 
all URIs within the RDF(S) and OWL namespace (and some additional). But in OWL 
1 DL, according to [2], there existed some URIs from these namespaces, which 
were actually allowed to be used in OWL DL ontologies. An example is the RDF 
reification vocabulary.

! Proposal: Explicitly itemize the disallowed URIs, just as in the old AS&S.

(B): Responsibility
-------------------

Technically, it doesn't matter whether the disallowed vocabulary is presented 
in the DL syntax or in the RDF mapping. However, there is a question of 
responsibility. I claim that having it in the DL syntax does not make much 
sense, while it there is a good reason to have it in the RDF mapping.

>From the point of view of OWL 2 DL and its Functional Syntax, there is no 
reason to /not/ allow, for example, the use of the URI 'rdfs:subClassOf' at, 
for example, any position within a class assertion. This URI is neither used 
within the Functional Syntax itself, nor does it have a special meaning in the 
DL semantics.

As an easy check, imagine that there was no RDF syntax for OWL DL. This would 
have no technical consequences for OWL DL, and the URIs from the RDF(S) and 
OWL vocabulary would then not be in any way special compared to any other URI. 
In fact, there would not even exist URIs in the OWL vocabulary.

An exception are probably a few URIs like 'owl:Thing' or 'rdfs:comment', which 
actually play a special role within the OWL 2 DL syntax. But then it would not 
make sense to disallow /every/ URI from the RDFS and OWL namespace. Instead, 
these few URIs could be treated specifically, or could be replaced by new 
names.

The only place where the disallowed vocabulary is really relevant is in the 
reverse RDF mapping. There, it might happen in some situations that the 
unrestricted usage of URIs from the RDF(S) or OWL vocabulary will lead to 
confusion when one applies the mapping rules, be it by making a reverse 
mapping impossible, or by leading to wrong results. This would then possibly 
break the "semantic roundtripping contract", which states that roundtripping 
from FS, through RDF and back to FS will always maintain the semantics of the 
original OWL DL ontology.

! Proposal: Move the definition of the disallowed vocabulary from the DL 
syntax document to the reverse RDF mapping.

(C) Relaxation
--------------

The argument in (B), that allowing certain RDF(S) or OWL vocabulary might lead 
to technical problems with the reverse RDF mapping, cannot be used to disallow 
/every/ URI from RDF(S) and OWL. Instead, it should really be checked for each 
such URI whether it leads to problems with the reverse mapping or not.

I had already some discussion on this topic for the concrete cases of rdf:List 
[3] and RDF reification [4] (though the use of the latter wasn't disallowed in 
OWL 1 DL, see (A)). And I think that in the analog way as discussed in these 
mails, several other URIs could be made accessible in OWL 2 DL.

One has to check this, of course, and this probably demands some effort. And 
another question will be which of these URIs are of interest for custom usage 
at all. However, as I have heard several times, there was a desire to make 
more RDF graphs valid in OWL 2 DL. So here is a real chance to get more valid 
OWL DL ontologies in RDF graph form!

Best,
Michael

[1] <http://www.w3.org/2007/OWL/wiki/Syntax#URIs_and_Namespaces>
[2] <http://www.w3.org/TR/owl-semantics/mapping.html#4.2>
[3] <http://lists.w3.org/Archives/Public/public-owl-wg/2008May/0188.html>
[4] <http://lists.w3.org/Archives/Public/public-owl-wg/2008May/0190.html>